> ## Documentation Index
> Fetch the complete documentation index at: https://noorle.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Adding Sources

> Upload files or paste text into a knowledge base, and follow ingestion to completion.

A knowledge base holds **sources**. Add them from the knowledge base's detail
page in the Portal.

## Two ways to add one

### Upload files

**Upload files** opens a multi-select file picker. Accepted extensions:

| Extension | Type |
| - | - |
| `.pdf` | PDF |
| `.docx` | Word |
| `.xlsx` | Excel |
| `.pptx` | PowerPoint |
| `.txt` | Plain text |
| `.md` | Markdown |
| `.csv` | CSV |
| `.html` / `.htm` | HTML |

**Limit: 50 MiB per file.** A larger file is rejected with a clear error rather
than truncated.

> [!NOTE]
>
> JSON is not an accepted type. Convert to CSV or Markdown first.

### Paste text

**Paste text** opens a drawer with a **Name** (up to 200 characters) and a
**Content** box. The Portal describes exactly what it does: "Stored as a markdown
document and indexed like any uploaded file."

Right for a policy, an FAQ, or a snippet that has no file of its own.

## Ingestion

Adding a source starts a pipeline: the file is stored, its text extracted, split
into chunks, embedded, and indexed. Each source shows a status while that runs
and the Portal polls until it settles.

A completed source reports how many chunks it produced. A failed one reports why.

Small text sources settle in seconds. Large PDFs take longer.

## Removing a source

Each row has a delete action. Removing a source removes its chunks from search.

## Knowledge base settings

The detail page also carries the knowledge base's own settings:

| Field | Meaning |
| - | - |
| **Name** | Shown to agents and teammates |
| **Description** | Optional — what lives in this knowledge base |
| **Max search results** | Cap on what this knowledge base contributes per query |

## Curating well

* **Text beats scans.** A PDF that is an image of text extracts poorly. Run OCR
  first, or paste the text.
* **Structure survives.** Headings and complete sentences chunk better than a
  wall of fragments.
* **Split enormous documents.** A 300-page manual as one source works; as five
  topic-scoped sources it retrieves better.
* **Delete what is stale.** Retrieval has no idea a document is out of date. An
  obsolete policy will be returned confidently.
* **Re-upload to update.** There is no in-place edit — delete the old source and
  add the new one.

## Next

* [Knowledge bases](https://noorle.com/docs/run/knowledge/overview.md)
* [Knowledge Retrieval capability](https://noorle.com/docs/run/capabilities/knowledge-retrieval.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.